text_classifier.py

from re import match
from spacy import load
from spacy.matcher import Matcher, PhraseMatcher
from spacy.tokenizer import Tokenizer
from spacy.tokens import Token
from dateparser import parse

# Auxiliary functions
def is_date(string):
    try:
        parse(str(string))
        return True

    except ValueError:
        return False

def is_email(string):
    return match(r'\S+@\S+', str(string))

def is_phone(string):
    return match(r'\d{8,10}', str(string))

def classify(text):
  nlp = load('es_core_news_sm')
  matcher = Matcher(nlp.vocab)

  # Dni number Pattern
  pattern = [{"TEXT": u'DNI'}, {"SHAPE": "dd.ddd.ddd"}]
  matcher.add("DNI", None, pattern)

    # Dni number Pattern
  pattern = [{"SHAPE": "dd.ddd.ddd"}]
  matcher.add("DNI-NRO", None, pattern)

  # Phone Number Pattern
  pattern = [{"TEXT": u'Tel'},{"SHAPE": "dddd"}]
  matcher.add("TELEFONO", None, pattern)

  pattern = [{"SHAPE": "dd"}, {"ORTH":"-"}, {"SHAPE": "dddd"}, {"ORTH":"-"}, {"SHAPE": "dddd"}]
  matcher.add("TELEFONO1", None, pattern)

  pattern = [{"TEXT": u'Florencia'}]
  matcher.add("FLORENCIA", None, pattern)

  contenido = text
  doc = nlp(contenido)
  res = []
  matches = matcher(doc)

  for match_id, start, end in matches:
      span = doc[start:end]
      res.append((nlp.vocab.strings[match_id], start, span.text))

  for ent in doc.ents:
      if (ent.label_ == 'PER' and not is_email(ent.text)):
          res.append(('NOMBRE', ent.start, ent.text))

      if (ent.label_ == 'LOC' and not is_email(ent) and not is_phone(ent)):
          res.append(('LUGAR', ent.start, ent.text))


  for token in doc:

      if (token.like_email):
          res.append(('EMAIL', token.idx, token.text))

      if (token.like_num and is_phone(token.text)):
          res.append(('TELEFONO', token.idx, token.text))

  return res